Year: 2025
Type: dissertation
Abstract: Artificial intelligence (AI) has made huge impact in cybercrime and cybersecurity over the recent years. While AI has enabled new forms of cyberattacks it has also provided newer and better performing AI powered cyber defense tools. This narrative review is aimed at reviewing the algorithms behind the prevalent AI anomaly detection tools and assessing their performance based on key metrics. This review employed a narrative approach followed by a comparative analysis. Two databases were used to search for the relevant literature. Studies were included based on the relevance to the research questions and the time of publication. Comparison of machine learning (ML) models based on key performance metrics such as accuracy, precision, recall and F1-Score was carried out. Fundamental concepts behind ML techniques, performance metrics, and their application within the domain of cloud security are discussed. Among selected AI-based anomaly and malware detection techniques the unsupervised learning based DBSCAN method delivered excellent performance, while deep learning methods showed significant improvement in identifying new and unknown attack patterns. Though supervised models had limitations in terms of false negative rates, they delivered better accuracy in detecting known anomaly patterns. Additionally, real-world AI driven anomaly detection tools such as the Microsoft Sentinel and the DeepLog have robust machine leaning capabilities and efficiency in combating cyberattacks. Deepfake detection tools, including Intel’s FakeCatcher and DeepFake-O-Meter, also delivered excellent accuracy in identifying fake media. ML models are well-suited for combating modern cyberattacks and integrating multiple ML models based on key performance metrics can further strengthen the cyber defense systems based on tailored needs.
Source: Theseus (Ammattikorkeakoulujen)
Author: Stephen, Godwin
Language: English
Subfield: Computer Networks and Communications
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7019812740
